Applications of Computational Intelligence

Speakers:

Using Bayesian Network Classifiers To Analyze Gene Expression Data: A Case Study

Jie Cheng, Siemens Corporate Research

This paper uses a well-studied public gene expression dataset to illustrate our approach to high dimensional data analysis using Bayesian network learning. Various techniques have been applied, such as feature selection, feature discretization, and model validation. This paper also presents the experimental results and discusses the lessons learned.

Using ROC Curve In the Absence of Positive Examples

Zehra Cataltepe, Ming-Wei Chang, Thomas Stiefmeier, Bo-Juen Chen and Claus Neubauer, Siemens Corporate Research

Receiver Operating Characteristics (ROC) curves have long been used to evaluate classifier performance in many fields (e.g. signal detection and machine learning). The ROC curve provides information on the tradeoff between the hit rate (true positives) and the false alarm rates (false positives). In order to draw the ROC curve both positive and negative examples are needed. In some applications, for example, machine condition monitoring, cancer detection, there are plenty of negative examples. However the positive examples are either rare, or do not fully describe the overall set of the possible positive examples. However, instead of the positive examples, some rules about the positive examples are available. For example, in machine condition monitoring, if a sensor drifts off from the set of observed states by a certain amount, we know that a fault has occurred. In order to use the ROC curve to evaluate classifiers, we artificially create the positive examples based on the application dependent rules and the existing negative examples. Then we draw the ROC curve using this set of positive and negative examples.

Image Watermarking Using Fuzzy Logic

Ayman M. Ahmed and Dwight D. Day, Kansas State University

In this paper an image watermarking approach using fuzzy inference system (FIS) is presented. The FIS extracts the most robust regions in the host image based on the relationships between pixels in the texture images extracted from the host images. The FIS output is multiplied by a logo image to redistribute logo in an invisible and robust way based on the distribution of the most robust textured regions in the host image. Experiential results are demonstrated to test the robustness of the proposed approach.

Image Watermarking Using Hartley Based Naturalness Preserving Transform

Ayman M. Ahmed and Dwight D. Day, Kansas State University

This paper proposes a new image-watermarking scheme. This approach uses the Naturalness Preserving Transform (NPT) to code a watermark into a host image. A new modified form of the NPT employing the Hartley transform is presented to address the artifacts effects that occur to images transformed by the original Hadamard based NPT. The modified Hartley based NPT enhances the quality of the transformed image in a sense that no noticeable difference will occur to distinguish between the transformed image and the original image. Low pass filtering and JPEG compression are selected to test the robustness of the proposed watermark.

Fast Pattern Recognition Incorporating Biological Principles

Claus Neubauer and M. Fang, Siemens Corporate Research

Algorithms for recognition of objects are interesting both from a computational intelligence point of view as well as for practical industrial applications. A new method for fast feature based 2D object recognition and localization in gray scale images is proposed, that relies on an oriented edge representation of the object. In the preprocessing stage sub-sampling, smoothing and edge filtering is performed. The edges are thinned and edge orientation and length of edge segments are extracted. Subsequently, a distance transformation is computed on the edge image. Finally, a fast matching process taking point distances as well as edge orientation into account reliably retrieves the position and orientation of multiple objects in the scene even in the presence of heavy clutter.

Last modified: June 18, 2003 by Dan Ventura.